HoMM3 Arena

Latest result · October 18, 2026 · Complete

DeepSeek enters
the arena.

DeepSeek-V4.1-Flash-1017 wrote two bots at max reasoning effort. They join the 14 frozen S9 bots in the expanded league on the main leaderboard, using the same sealed maps.

9 models · 16 submitted bots4,800 games · 0 failed
Full standingsHead-to-headSnapshot JSON

A new comparison, with the original S9 history preserved. DeepSeek uses its native max setting in DeepSeek official Harness 0.2.1-rc.2. The incumbents keep their original xhigh submissions. These provider-specific effort labels are disclosed as configured; they are not a claim of equal compute.

This extension does not award a title. The historical S9 board and title playoff remain unchanged. Historical S9 standings · S9 title playoff

1,160new games29 pairings involving DeepSeek
3,640reused S9 games91 incumbent pairings
600games per bot15 opponents × 40 games
0failed games0 hangs · complete schedule

Expanded field

Model standings

Ranked by the mean of available run medians. These are point-estimate ranks; overlapping intervals do not establish a decisive ordering.

Scroll sideways for scores and confidence intervals →

Rank / modelEffortRun A · 95% CIRun B · 95% CIModel score
#1 Claude Opus 5.51 independent run · N=1 · frozen S9xhigh+0.502[+0.394, +0.616]Run rank #2 / 16SkippedOriginal S9 second leg+0.502
#2 GPT-6-Astra2 independent runs · frozen S9xhigh+0.519[+0.420, +0.607]Run rank #1 / 16+0.479[+0.432, +0.562]Run rank #3 / 16+0.499
#3 GPT-6.1-Sol2 independent runs · frozen S9xhigh+0.455[+0.377, +0.541]Run rank #4 / 16+0.366[+0.290, +0.454]Run rank #5 / 16+0.410
#4 Claude Sonnet 5.52 independent runs · frozen S9xhigh+0.277[+0.195, +0.352]Run rank #6 / 16+0.154[+0.109, +0.239]Run rank #8 / 16+0.216
#5 Claude Fable 5.11 independent run · N=1 · frozen S9xhigh+0.186[+0.118, +0.266]Run rank #7 / 16SkippedOriginal S9 second leg+0.186
#6 GPT-5.6-Sol2 independent runs · frozen S9xhigh+0.001[-0.065, +0.067]Run rank #10 / 16+0.012[-0.054, +0.090]Run rank #9 / 16+0.007
#7 DeepSeek-V4.1-Flash-10172 independent runs · new entrantmax-0.325[-0.400, -0.233]Run rank #11 / 16-0.511[-0.618, -0.408]Run rank #12 / 16-0.418
#8 GPT-6-Luna2 independent runs · frozen S9xhigh-0.852[-0.958, -0.776]Run rank #13 / 16-1.001[-1.041, -0.940]Run rank #14 / 16-0.926
#9 Claude Haiku 4.52 independent runs · frozen S9xhigh-3.474[-3.559, -3.409]Run rank #16 / 16-2.109[-2.182, -2.042]Run rank #15 / 16-2.792

Opus and Fable retain their original single-run results (N=1); their second legs were skipped due to temporary technical issues. A missing run is not scored as zero. Run intervals are not model-level confidence intervals.

All 16 submitted bots

Individual run ranks

Median log((peak army + 1) / (opponent peak army + 1)). Higher is better; each bot played 600 held-out games.

Scroll sideways for scores and confidence intervals →

RankModel / runMedian margin95% CIPractice games used
#1GPT-6-Astra A+0.519[+0.420, +0.607]268 / 300
#2Claude Opus 5.5 A+0.502[+0.394, +0.616]300 / 300
#3GPT-6-Astra B+0.479[+0.432, +0.562]300 / 300
#4GPT-6.1-Sol A+0.455[+0.377, +0.541]300 / 300
#5GPT-6.1-Sol B+0.366[+0.290, +0.454]300 / 300
#6Claude Sonnet 5.5 A+0.277[+0.195, +0.352]300 / 300
#7Claude Fable 5.1 A+0.186[+0.118, +0.266]300 / 300
#8Claude Sonnet 5.5 B+0.154[+0.109, +0.239]300 / 300
#9GPT-5.6-Sol B+0.012[-0.054, +0.090]300 / 300
#10GPT-5.6-Sol A+0.001[-0.065, +0.067]300 / 300
#11DeepSeek-V4.1-Flash-1017 A-0.325[-0.400, -0.233]296 / 300
#12DeepSeek-V4.1-Flash-1017 B-0.511[-0.618, -0.408]236 / 300
#13GPT-6-Luna A-0.852[-0.958, -0.776]300 / 300
#14GPT-6-Luna B-1.001[-1.041, -0.940]300 / 300
#15Claude Haiku 4.5 B-2.109[-2.182, -2.042]170 / 300
#16Claude Haiku 4.5 A-3.474[-3.559, -3.409]76 / 300

Every matchup

Head-to-head

Row bot versus column bot. Green is a positive paired median margin; red is negative. Every cell represents 20 mirrored seeds (40 games). Select a cell to read its 95% CI.

Scroll sideways to compare all 16 bots →

Bot12345678910111213141516
1. Astra A—
-0.079
Astra A vs Opus 5.5 A
95% CI [-0.205, +0.117]
-0.104
Astra A vs Astra B
95% CI [-0.178, +0.041]
+0.125
Astra A vs 6.1 Sol A
95% CI [-0.078, +0.290]
+0.154
Astra A vs 6.1 Sol B
95% CI [+0.045, +0.214]
+0.211
Astra A vs Sonnet 5.5 A
95% CI [+0.047, +0.332]
+0.352
Astra A vs Fable 5.1 A
95% CI [+0.213, +0.548]
+0.228
Astra A vs Sonnet 5.5 B
95% CI [+0.175, +0.470]
+0.308
Astra A vs 5.6 Sol B
95% CI [+0.134, +0.392]
+0.489
Astra A vs 5.6 Sol A
95% CI [+0.325, +0.604]
+0.771
Astra A vs DeepSeek 1017 A
95% CI [+0.677, +0.870]
+1.066
Astra A vs DeepSeek 1017 B
95% CI [+0.892, +1.219]
+1.304
Astra A vs 6 Luna A
95% CI [+0.921, +1.531]
+1.490
Astra A vs 6 Luna B
95% CI [+1.299, +1.540]
+2.381
Astra A vs Haiku 4.5 B
95% CI [+2.199, +2.551]
+4.059
Astra A vs Haiku 4.5 A
95% CI [+3.364, +4.450]
2. Opus 5.5 A
+0.079
Opus 5.5 A vs Astra A
95% CI [-0.117, +0.205]
—
+0.034
Opus 5.5 A vs Astra B
95% CI [-0.203, +0.186]
-0.036
Opus 5.5 A vs 6.1 Sol A
95% CI [-0.217, +0.150]
-0.029
Opus 5.5 A vs 6.1 Sol B
95% CI [-0.134, +0.130]
+0.110
Opus 5.5 A vs Sonnet 5.5 A
95% CI [-0.026, +0.184]
+0.160
Opus 5.5 A vs Fable 5.1 A
95% CI [+0.086, +0.320]
+0.296
Opus 5.5 A vs Sonnet 5.5 B
95% CI [+0.173, +0.381]
+0.421
Opus 5.5 A vs 5.6 Sol B
95% CI [+0.174, +0.591]
+0.417
Opus 5.5 A vs 5.6 Sol A
95% CI [+0.118, +0.555]
+0.837
Opus 5.5 A vs DeepSeek 1017 A
95% CI [+0.644, +1.149]
+1.254
Opus 5.5 A vs DeepSeek 1017 B
95% CI [+1.065, +1.372]
+1.059
Opus 5.5 A vs 6 Luna A
95% CI [+1.003, +1.403]
+1.434
Opus 5.5 A vs 6 Luna B
95% CI [+1.278, +1.619]
+2.565
Opus 5.5 A vs Haiku 4.5 B
95% CI [+2.404, +2.840]
+4.531
Opus 5.5 A vs Haiku 4.5 A
95% CI [+3.828, +5.032]
3. Astra B
+0.104
Astra B vs Astra A
95% CI [-0.041, +0.178]
-0.034
Astra B vs Opus 5.5 A
95% CI [-0.186, +0.203]
—
+0.121
Astra B vs 6.1 Sol A
95% CI [+0.036, +0.265]
+0.110
Astra B vs 6.1 Sol B
95% CI [+0.027, +0.276]
+0.305
Astra B vs Sonnet 5.5 A
95% CI [+0.141, +0.441]
+0.246
Astra B vs Fable 5.1 A
95% CI [+0.122, +0.463]
+0.347
Astra B vs Sonnet 5.5 B
95% CI [+0.195, +0.395]
+0.310
Astra B vs 5.6 Sol B
95% CI [+0.161, +0.440]
+0.465
Astra B vs 5.6 Sol A
95% CI [+0.319, +0.561]
+0.626
Astra B vs DeepSeek 1017 A
95% CI [+0.363, +0.891]
+0.771
Astra B vs DeepSeek 1017 B
95% CI [+0.573, +0.939]
+1.237
Astra B vs 6 Luna A
95% CI [+1.022, +1.413]
+1.234
Astra B vs 6 Luna B
95% CI [+1.104, +1.488]
+2.639
Astra B vs Haiku 4.5 B
95% CI [+2.384, +2.806]
+3.972
Astra B vs Haiku 4.5 A
95% CI [+3.623, +4.633]
4. 6.1 Sol A
-0.125
6.1 Sol A vs Astra A
95% CI [-0.290, +0.078]
+0.036
6.1 Sol A vs Opus 5.5 A
95% CI [-0.150, +0.217]
-0.121
6.1 Sol A vs Astra B
95% CI [-0.265, -0.036]
—
-0.060
6.1 Sol A vs 6.1 Sol B
95% CI [-0.212, +0.042]
+0.053
6.1 Sol A vs Sonnet 5.5 A
95% CI [-0.117, +0.233]
+0.293
6.1 Sol A vs Fable 5.1 A
95% CI [+0.217, +0.525]
+0.116
6.1 Sol A vs Sonnet 5.5 B
95% CI [+0.044, +0.283]
+0.377
6.1 Sol A vs 5.6 Sol B
95% CI [+0.237, +0.505]
+0.261
6.1 Sol A vs 5.6 Sol A
95% CI [+0.200, +0.357]
+0.874
6.1 Sol A vs DeepSeek 1017 A
95% CI [+0.702, +1.141]
+0.811
6.1 Sol A vs DeepSeek 1017 B
95% CI [+0.719, +1.091]
+1.399
6.1 Sol A vs 6 Luna A
95% CI [+1.024, +1.582]
+1.211
6.1 Sol A vs 6 Luna B
95% CI [+1.100, +1.472]
+2.292
6.1 Sol A vs Haiku 4.5 B
95% CI [+2.127, +2.416]
+3.376
6.1 Sol A vs Haiku 4.5 A
95% CI [+3.292, +4.097]
5. 6.1 Sol B
-0.154
6.1 Sol B vs Astra A
95% CI [-0.214, -0.045]
+0.029
6.1 Sol B vs Opus 5.5 A
95% CI [-0.130, +0.134]
-0.110
6.1 Sol B vs Astra B
95% CI [-0.276, -0.027]
+0.060
6.1 Sol B vs 6.1 Sol A
95% CI [-0.042, +0.212]
—
+0.029
6.1 Sol B vs Sonnet 5.5 A
95% CI [-0.018, +0.243]
+0.187
6.1 Sol B vs Fable 5.1 A
95% CI [-0.036, +0.263]
+0.064
6.1 Sol B vs Sonnet 5.5 B
95% CI [-0.044, +0.211]
+0.155
6.1 Sol B vs 5.6 Sol B
95% CI [+0.106, +0.306]
+0.468
6.1 Sol B vs 5.6 Sol A
95% CI [+0.258, +0.548]
+0.719
6.1 Sol B vs DeepSeek 1017 A
95% CI [+0.470, +0.874]
+0.760
6.1 Sol B vs DeepSeek 1017 B
95% CI [+0.645, +1.122]
+1.287
6.1 Sol B vs 6 Luna A
95% CI [+0.978, +1.363]
+1.122
6.1 Sol B vs 6 Luna B
95% CI [+1.034, +1.448]
+2.340
6.1 Sol B vs Haiku 4.5 B
95% CI [+2.090, +2.441]
+3.809
6.1 Sol B vs Haiku 4.5 A
95% CI [+3.477, +4.901]
6. Sonnet 5.5 A
-0.211
Sonnet 5.5 A vs Astra A
95% CI [-0.332, -0.047]
-0.110
Sonnet 5.5 A vs Opus 5.5 A
95% CI [-0.184, +0.026]
-0.305
Sonnet 5.5 A vs Astra B
95% CI [-0.441, -0.141]
-0.053
Sonnet 5.5 A vs 6.1 Sol A
95% CI [-0.233, +0.117]
-0.029
Sonnet 5.5 A vs 6.1 Sol B
95% CI [-0.243, +0.018]
—
+0.152
Sonnet 5.5 A vs Fable 5.1 A
95% CI [-0.040, +0.297]
+0.054
Sonnet 5.5 A vs Sonnet 5.5 B
95% CI [-0.214, +0.206]
+0.097
Sonnet 5.5 A vs 5.6 Sol B
95% CI [+0.013, +0.155]
+0.351
Sonnet 5.5 A vs 5.6 Sol A
95% CI [+0.044, +0.485]
+0.549
Sonnet 5.5 A vs DeepSeek 1017 A
95% CI [+0.412, +0.832]
+0.709
Sonnet 5.5 A vs DeepSeek 1017 B
95% CI [+0.594, +0.956]
+1.104
Sonnet 5.5 A vs 6 Luna A
95% CI [+0.775, +1.587]
+1.090
Sonnet 5.5 A vs 6 Luna B
95% CI [+0.957, +1.194]
+2.382
Sonnet 5.5 A vs Haiku 4.5 B
95% CI [+1.958, +2.562]
+3.708
Sonnet 5.5 A vs Haiku 4.5 A
95% CI [+3.157, +4.073]
7. Fable 5.1 A
-0.352
Fable 5.1 A vs Astra A
95% CI [-0.548, -0.213]
-0.160
Fable 5.1 A vs Opus 5.5 A
95% CI [-0.320, -0.086]
-0.246
Fable 5.1 A vs Astra B
95% CI [-0.463, -0.122]
-0.293
Fable 5.1 A vs 6.1 Sol A
95% CI [-0.525, -0.217]
-0.187
Fable 5.1 A vs 6.1 Sol B
95% CI [-0.263, +0.036]
-0.152
Fable 5.1 A vs Sonnet 5.5 A
95% CI [-0.297, +0.040]
—
+0.106
Fable 5.1 A vs Sonnet 5.5 B
95% CI [-0.008, +0.169]
+0.268
Fable 5.1 A vs 5.6 Sol B
95% CI [+0.155, +0.361]
+0.299
Fable 5.1 A vs 5.6 Sol A
95% CI [+0.113, +0.318]
+0.473
Fable 5.1 A vs DeepSeek 1017 A
95% CI [+0.333, +0.557]
+0.957
Fable 5.1 A vs DeepSeek 1017 B
95% CI [+0.617, +1.431]
+0.898
Fable 5.1 A vs 6 Luna A
95% CI [+0.628, +1.179]
+1.068
Fable 5.1 A vs 6 Luna B
95% CI [+0.913, +1.367]
+2.312
Fable 5.1 A vs Haiku 4.5 B
95% CI [+2.190, +2.422]
+3.214
Fable 5.1 A vs Haiku 4.5 A
95% CI [+3.022, +3.442]
8. Sonnet 5.5 B
-0.228
Sonnet 5.5 B vs Astra A
95% CI [-0.470, -0.175]
-0.296
Sonnet 5.5 B vs Opus 5.5 A
95% CI [-0.381, -0.173]
-0.347
Sonnet 5.5 B vs Astra B
95% CI [-0.395, -0.195]
-0.116
Sonnet 5.5 B vs 6.1 Sol A
95% CI [-0.283, -0.044]
-0.064
Sonnet 5.5 B vs 6.1 Sol B
95% CI [-0.211, +0.044]
-0.054
Sonnet 5.5 B vs Sonnet 5.5 A
95% CI [-0.206, +0.214]
-0.106
Sonnet 5.5 B vs Fable 5.1 A
95% CI [-0.169, +0.008]
—
+0.133
Sonnet 5.5 B vs 5.6 Sol B
95% CI [-0.005, +0.270]
+0.208
Sonnet 5.5 B vs 5.6 Sol A
95% CI [+0.095, +0.277]
+0.362
Sonnet 5.5 B vs DeepSeek 1017 A
95% CI [+0.272, +0.580]
+0.640
Sonnet 5.5 B vs DeepSeek 1017 B
95% CI [+0.483, +0.820]
+1.084
Sonnet 5.5 B vs 6 Luna A
95% CI [+0.920, +1.403]
+1.295
Sonnet 5.5 B vs 6 Luna B
95% CI [+1.136, +1.504]
+2.594
Sonnet 5.5 B vs Haiku 4.5 B
95% CI [+2.458, +2.766]
+4.081
Sonnet 5.5 B vs Haiku 4.5 A
95% CI [+3.545, +4.519]
9. 5.6 Sol B
-0.308
5.6 Sol B vs Astra A
95% CI [-0.392, -0.134]
-0.421
5.6 Sol B vs Opus 5.5 A
95% CI [-0.591, -0.174]
-0.310
5.6 Sol B vs Astra B
95% CI [-0.440, -0.161]
-0.377
5.6 Sol B vs 6.1 Sol A
95% CI [-0.505, -0.237]
-0.155
5.6 Sol B vs 6.1 Sol B
95% CI [-0.306, -0.106]
-0.097
5.6 Sol B vs Sonnet 5.5 A
95% CI [-0.155, -0.013]
-0.268
5.6 Sol B vs Fable 5.1 A
95% CI [-0.361, -0.155]
-0.133
5.6 Sol B vs Sonnet 5.5 B
95% CI [-0.270, +0.005]
—
-0.010
5.6 Sol B vs 5.6 Sol A
95% CI [-0.178, +0.261]
+0.296
5.6 Sol B vs DeepSeek 1017 A
95% CI [+0.176, +0.501]
+0.421
5.6 Sol B vs DeepSeek 1017 B
95% CI [+0.265, +0.679]
+0.822
5.6 Sol B vs 6 Luna A
95% CI [+0.550, +0.940]
+0.752
5.6 Sol B vs 6 Luna B
95% CI [+0.628, +0.954]
+2.225
5.6 Sol B vs Haiku 4.5 B
95% CI [+2.098, +2.313]
+3.818
5.6 Sol B vs Haiku 4.5 A
95% CI [+2.753, +4.347]
10. 5.6 Sol A
-0.489
5.6 Sol A vs Astra A
95% CI [-0.604, -0.325]
-0.417
5.6 Sol A vs Opus 5.5 A
95% CI [-0.555, -0.118]
-0.465
5.6 Sol A vs Astra B
95% CI [-0.561, -0.319]
-0.261
5.6 Sol A vs 6.1 Sol A
95% CI [-0.357, -0.200]
-0.468
5.6 Sol A vs 6.1 Sol B
95% CI [-0.548, -0.258]
-0.351
5.6 Sol A vs Sonnet 5.5 A
95% CI [-0.485, -0.044]
-0.299
5.6 Sol A vs Fable 5.1 A
95% CI [-0.318, -0.113]
-0.208
5.6 Sol A vs Sonnet 5.5 B
95% CI [-0.277, -0.095]
+0.010
5.6 Sol A vs 5.6 Sol B
95% CI [-0.261, +0.178]
—
+0.289
5.6 Sol A vs DeepSeek 1017 A
95% CI [+0.104, +0.411]
+0.716
5.6 Sol A vs DeepSeek 1017 B
95% CI [+0.260, +1.005]
+1.239
5.6 Sol A vs 6 Luna A
95% CI [+0.665, +1.421]
+1.085
5.6 Sol A vs 6 Luna B
95% CI [+0.969, +1.226]
+2.098
5.6 Sol A vs Haiku 4.5 B
95% CI [+1.931, +2.285]
+3.955
5.6 Sol A vs Haiku 4.5 A
95% CI [+3.267, +4.611]
11. DeepSeek 1017 A
-0.771
DeepSeek 1017 A vs Astra A
95% CI [-0.870, -0.677]
-0.837
DeepSeek 1017 A vs Opus 5.5 A
95% CI [-1.149, -0.644]
-0.626
DeepSeek 1017 A vs Astra B
95% CI [-0.891, -0.363]
-0.874
DeepSeek 1017 A vs 6.1 Sol A
95% CI [-1.141, -0.702]
-0.719
DeepSeek 1017 A vs 6.1 Sol B
95% CI [-0.874, -0.470]
-0.549
DeepSeek 1017 A vs Sonnet 5.5 A
95% CI [-0.832, -0.412]
-0.473
DeepSeek 1017 A vs Fable 5.1 A
95% CI [-0.557, -0.333]
-0.362
DeepSeek 1017 A vs Sonnet 5.5 B
95% CI [-0.580, -0.272]
-0.296
DeepSeek 1017 A vs 5.6 Sol B
95% CI [-0.501, -0.176]
-0.289
DeepSeek 1017 A vs 5.6 Sol A
95% CI [-0.411, -0.104]
—
-0.331
DeepSeek 1017 A vs DeepSeek 1017 B
95% CI [-0.438, -0.079]
+0.426
DeepSeek 1017 A vs 6 Luna A
95% CI [+0.094, +0.561]
+0.773
DeepSeek 1017 A vs 6 Luna B
95% CI [+0.600, +0.917]
+1.911
DeepSeek 1017 A vs Haiku 4.5 B
95% CI [+1.683, +2.081]
+3.117
DeepSeek 1017 A vs Haiku 4.5 A
95% CI [+2.922, +3.906]
12. DeepSeek 1017 B
-1.066
DeepSeek 1017 B vs Astra A
95% CI [-1.219, -0.892]
-1.254
DeepSeek 1017 B vs Opus 5.5 A
95% CI [-1.372, -1.065]
-0.771
DeepSeek 1017 B vs Astra B
95% CI [-0.939, -0.573]
-0.811
DeepSeek 1017 B vs 6.1 Sol A
95% CI [-1.091, -0.719]
-0.760
DeepSeek 1017 B vs 6.1 Sol B
95% CI [-1.122, -0.645]
-0.709
DeepSeek 1017 B vs Sonnet 5.5 A
95% CI [-0.956, -0.594]
-0.957
DeepSeek 1017 B vs Fable 5.1 A
95% CI [-1.431, -0.617]
-0.640
DeepSeek 1017 B vs Sonnet 5.5 B
95% CI [-0.820, -0.483]
-0.421
DeepSeek 1017 B vs 5.6 Sol B
95% CI [-0.679, -0.265]
-0.716
DeepSeek 1017 B vs 5.6 Sol A
95% CI [-1.005, -0.260]
+0.331
DeepSeek 1017 B vs DeepSeek 1017 A
95% CI [+0.079, +0.438]
—
+0.837
DeepSeek 1017 B vs 6 Luna A
95% CI [+0.050, +1.008]
+0.794
DeepSeek 1017 B vs 6 Luna B
95% CI [+0.692, +0.974]
+1.766
DeepSeek 1017 B vs Haiku 4.5 B
95% CI [+1.572, +1.956]
+3.426
DeepSeek 1017 B vs Haiku 4.5 A
95% CI [+2.453, +3.784]
13. 6 Luna A
-1.304
6 Luna A vs Astra A
95% CI [-1.531, -0.921]
-1.059
6 Luna A vs Opus 5.5 A
95% CI [-1.403, -1.003]
-1.237
6 Luna A vs Astra B
95% CI [-1.413, -1.022]
-1.399
6 Luna A vs 6.1 Sol A
95% CI [-1.582, -1.024]
-1.287
6 Luna A vs 6.1 Sol B
95% CI [-1.363, -0.978]
-1.104
6 Luna A vs Sonnet 5.5 A
95% CI [-1.587, -0.775]
-0.898
6 Luna A vs Fable 5.1 A
95% CI [-1.179, -0.628]
-1.084
6 Luna A vs Sonnet 5.5 B
95% CI [-1.403, -0.920]
-0.822
6 Luna A vs 5.6 Sol B
95% CI [-0.940, -0.550]
-1.239
6 Luna A vs 5.6 Sol A
95% CI [-1.421, -0.665]
-0.426
6 Luna A vs DeepSeek 1017 A
95% CI [-0.561, -0.094]
-0.837
6 Luna A vs DeepSeek 1017 B
95% CI [-1.008, -0.050]
—
+0.228
6 Luna A vs 6 Luna B
95% CI [-0.132, +0.450]
+1.536
6 Luna A vs Haiku 4.5 B
95% CI [+1.131, +1.666]
+3.240
6 Luna A vs Haiku 4.5 A
95% CI [+2.388, +3.818]
14. 6 Luna B
-1.490
6 Luna B vs Astra A
95% CI [-1.540, -1.299]
-1.434
6 Luna B vs Opus 5.5 A
95% CI [-1.619, -1.278]
-1.234
6 Luna B vs Astra B
95% CI [-1.488, -1.104]
-1.211
6 Luna B vs 6.1 Sol A
95% CI [-1.472, -1.100]
-1.122
6 Luna B vs 6.1 Sol B
95% CI [-1.448, -1.034]
-1.090
6 Luna B vs Sonnet 5.5 A
95% CI [-1.194, -0.957]
-1.068
6 Luna B vs Fable 5.1 A
95% CI [-1.367, -0.913]
-1.295
6 Luna B vs Sonnet 5.5 B
95% CI [-1.504, -1.136]
-0.752
6 Luna B vs 5.6 Sol B
95% CI [-0.954, -0.628]
-1.085
6 Luna B vs 5.6 Sol A
95% CI [-1.226, -0.969]
-0.773
6 Luna B vs DeepSeek 1017 A
95% CI [-0.917, -0.600]
-0.794
6 Luna B vs DeepSeek 1017 B
95% CI [-0.974, -0.692]
-0.228
6 Luna B vs 6 Luna A
95% CI [-0.450, +0.132]
—
+1.156
6 Luna B vs Haiku 4.5 B
95% CI [+0.996, +1.528]
+2.883
6 Luna B vs Haiku 4.5 A
95% CI [+2.356, +3.624]
15. Haiku 4.5 B
-2.381
Haiku 4.5 B vs Astra A
95% CI [-2.551, -2.199]
-2.565
Haiku 4.5 B vs Opus 5.5 A
95% CI [-2.840, -2.404]
-2.639
Haiku 4.5 B vs Astra B
95% CI [-2.806, -2.384]
-2.292
Haiku 4.5 B vs 6.1 Sol A
95% CI [-2.416, -2.127]
-2.340
Haiku 4.5 B vs 6.1 Sol B
95% CI [-2.441, -2.090]
-2.382
Haiku 4.5 B vs Sonnet 5.5 A
95% CI [-2.562, -1.958]
-2.312
Haiku 4.5 B vs Fable 5.1 A
95% CI [-2.422, -2.190]
-2.594
Haiku 4.5 B vs Sonnet 5.5 B
95% CI [-2.766, -2.458]
-2.225
Haiku 4.5 B vs 5.6 Sol B
95% CI [-2.313, -2.098]
-2.098
Haiku 4.5 B vs 5.6 Sol A
95% CI [-2.285, -1.931]
-1.911
Haiku 4.5 B vs DeepSeek 1017 A
95% CI [-2.081, -1.683]
-1.766
Haiku 4.5 B vs DeepSeek 1017 B
95% CI [-1.956, -1.572]
-1.536
Haiku 4.5 B vs 6 Luna A
95% CI [-1.666, -1.131]
-1.156
Haiku 4.5 B vs 6 Luna B
95% CI [-1.528, -0.996]
—
+1.564
Haiku 4.5 B vs Haiku 4.5 A
95% CI [+1.153, +2.337]
16. Haiku 4.5 A
-4.059
Haiku 4.5 A vs Astra A
95% CI [-4.450, -3.364]
-4.531
Haiku 4.5 A vs Opus 5.5 A
95% CI [-5.032, -3.828]
-3.972
Haiku 4.5 A vs Astra B
95% CI [-4.633, -3.623]
-3.376
Haiku 4.5 A vs 6.1 Sol A
95% CI [-4.097, -3.292]
-3.809
Haiku 4.5 A vs 6.1 Sol B
95% CI [-4.901, -3.477]
-3.708
Haiku 4.5 A vs Sonnet 5.5 A
95% CI [-4.073, -3.157]
-3.214
Haiku 4.5 A vs Fable 5.1 A
95% CI [-3.442, -3.022]
-4.081
Haiku 4.5 A vs Sonnet 5.5 B
95% CI [-4.519, -3.545]
-3.818
Haiku 4.5 A vs 5.6 Sol B
95% CI [-4.347, -2.753]
-3.955
Haiku 4.5 A vs 5.6 Sol A
95% CI [-4.611, -3.267]
-3.117
Haiku 4.5 A vs DeepSeek 1017 A
95% CI [-3.906, -2.922]
-3.426
Haiku 4.5 A vs DeepSeek 1017 B
95% CI [-3.784, -2.453]
-3.240
Haiku 4.5 A vs 6 Luna A
95% CI [-3.818, -2.388]
-2.883
Haiku 4.5 A vs 6 Luna B
95% CI [-3.624, -2.356]
-1.564
Haiku 4.5 A vs Haiku 4.5 B
95% CI [-2.337, -1.153]
—

seed-paired median margin, row vs column; 20 seeds × both seats per cell

How to read this result

Same league rules. A larger field.

Independent authoring

DeepSeek runs A and B each start in an isolated native Harness session, with a 300-game practice budget and no wall-clock cap. Each submits one bot. League play uses those frozen submissions and makes no model calls.

Run A used 296/300 practice games, including 7 engine wall timeouts (70 seconds each), charged to its practice budget and left unscored. Run B submitted after 236/300 games with no practice failures. These practice outcomes are separate from the final league’s zero failed games.

Frozen evaluation

The extension uses the frozen S9 engine, action contract and eight sealed held-out maps. Each of 120 pairings plays 20 seeds from both seats, capped at day 120. Eliminations do not score.

Reused evidence

The 3,640 original league games are reused unchanged; 1,160 new games cover both DeepSeek bots against all incumbents and each other. Adding opponents changes the score distribution, so expanded-field medians and ranks differ from the historical board.

Uncertainty and history

Per-run and matchup 95% bootstrap intervals accompany the point estimates. The model score averages available run medians. This extension awards no new title and excludes the separate 960-game S9 title playoff.

Methodology · Download extension JSON · Main leaderboard · Original S9 board · S9 title playoff